Opened 7 years ago

Closed 7 years ago

Last modified 7 years ago

#54295 closed enhancement (wontfix)

mumps @5.0.1: does not generate mpiseq library

Reported by: lnmaurer (Leon Nathaniel Maurer) Owned by: seanfarley (Sean Farley)
Priority: Normal Milestone:
Component: ports Version:
Keywords: Cc:
Port: mumps

Description

I've compiled mumps 5.0.2 on linux, and one of libraries it generated was libmpiseq.a, which allows mumps to be used in sequential (i.e. not parallel) mode. However, as far as I can tell, the mumps 5.0.1 MacPorts package does not generate the mpiseq library.

(MacPorts does have an older package named mumps_seq, which does generate the mpiseq library. However, dependency conflicts mean that you cannot have the mumps_seq and mumps packages installed at the same time.)

All this is a problem for me because kwant will only work with the sequential version of mumps; it requires the mpiseq library. (I can build kwant without issue when the mumps_seq package is installed.)

So, it would be nice for the mumps package to generate libmpiseq.a so that I don't need to use the older mumps_seq package.

(I should note that this report may be related to #40039, which concerns an older version of mumps.)

Change History (5)

comment:1 Changed 7 years ago by mf2k (Frank Schima)

Owner: set to seanfarley
Status: newassigned
Version: 2.4.1

In the future, please Cc the port maintainers (port info --maintainers mumps), if any.

comment:2 Changed 7 years ago by lnmaurer (Leon Nathaniel Maurer)

Will do!

comment:3 Changed 7 years ago by seanfarley (Sean Farley)

Status: assignedaccepted

Hmm, I seem to remember a problem with mpiseq ... lemme look into this and see what I can dig up.

comment:4 Changed 7 years ago by seanfarley (Sean Farley)

Resolution: wontfix
Status: acceptedclosed

Providing libmpiseq in MUMPS directly conflicts with other packages (namely, PETSc). Also, the MUMPS developers can't even namespace their own package correctly so it'd be a colossal maintenance nightmare to provide both. I wish the kwant developers would interface with a better linear algebra package (unsurprisingly, I recommend PETsc). Unless the condition number of these matrices are so bad that they need a direct solver (no idea off the top of my head). If dense matrices are preferred then there's Elemental.

I did try to patch kwant's setup.py but it took more time that I currently have to see if they could be cajoled into using parallel MUMPS (their reason of singled core-ness is dubious to me). Anything, literally anything, would be better than writing an interface to MUMPS.

I'm going to close this as wontfix because I simply don't have the time to maintain another MUMPS package. If someone wants to step up, I could provide the work I currently have to make a subport of MUMPS that is sequential only.

comment:5 Changed 7 years ago by lnmaurer (Leon Nathaniel Maurer)

Roger that. Thanks for taking a look.

Note: See TracTickets for help on using tickets.